Re: [2.6.16-gitX] heavy performance regression in ipw2200 wireless driver

Re: [2.6.16-gitX] heavy performance regression in ipw2200 wireless driver

[Andrew Morton]

"Alessandro Suardi" <alessandro.suardi@gmail.com> wrote:
>

Pleeeeze try to cc the right people.

> Driver - or firmware ? Don't know - since the new git snapshots run
>  1.1.1 which requires newer firmware from http://ipw2200.sourceforge.net.
> 
> Symptom -> my new FC5 partition with 2.6.16-git kernels connects via
>  VNC viewer to my bittorrent box over wireless (ipw2200 to a D-Link
>  G604T router/AP); Dell D610 runs FC5, BT box is a K7-800 running
>  FC3 with a 2.6.16-rc5-git8 kernel (15+ days uptime...).
> 
> I also run Firefox on the bittorrent box; noticed today (2.6.16-git5) that
>  the screen refresh of pages with images was from time to time very
>  slow (close to unusable).
> 
> Rebooted into my FC4 partition with a 2.6.16 kernel, everything much
>  snappier. So I ran a scp test from my BT server to the laptop, three
>  times in a row the same file - a 38MB .flac with the laptop in the same
>  physical position (ie, no signal variation). Results...
> 
> FC5 - 2.6.16-git3:
> 
> [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> asuardi@192.168.1.8's password:
> KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB 971.3KB/s   00:40
> [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> asuardi@192.168.1.8's password:
> KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB   1.3MB/s   00:29
> [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> asuardi@192.168.1.8's password:
> KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB 626.7KB/s   01:02
> 
> 
> FC4 - 2.6.16:
> 
> [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> asuardi@192.168.1.8's password:
> KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB   1.5MB/s   00:25
> [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> asuardi@192.168.1.8's password:
> KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB   1.7MB/s   00:23
> [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> asuardi@192.168.1.8's password:
> KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB   1.7MB/s   00:22
> 
> Bottom line - old driver has better performance than the new one,
>  but most noticeably delivers consistent performance.
> 
> I will be available for testing starting Thursday 30th as I'll be on
>  the road since then. Of course if the problem is identified and
>  fixed earlier, I won't cry ;)

Well.  It's not a huge regression.  It's a 50%ish regression.  We've done
worse ;)

Re: [2.6.16-gitX] heavy performance regression in ipw2200 wireless driver

[Alessandro Suardi]

On 3/23/06, Andrew Morton <akpm@osdl.org> wrote:
> "Alessandro Suardi" <alessandro.suardi@gmail.com> wrote:
> >
>
> Pleeeeze try to cc the right people.

Sorry about that - should probably defer bug reporting to times
 when I'm actually supposed to be awake (2:20am doesn't fit
 the bill obviously :| )

> > Driver - or firmware ? Don't know - since the new git snapshots run
> >  1.1.1 which requires newer firmware from http://ipw2200.sourceforge.net.
> >
> > Symptom -> my new FC5 partition with 2.6.16-git kernels connects via
> >  VNC viewer to my bittorrent box over wireless (ipw2200 to a D-Link
> >  G604T router/AP); Dell D610 runs FC5, BT box is a K7-800 running
> >  FC3 with a 2.6.16-rc5-git8 kernel (15+ days uptime...).
> >
> > I also run Firefox on the bittorrent box; noticed today (2.6.16-git5) that
> >  the screen refresh of pages with images was from time to time very
> >  slow (close to unusable).
> >
> > Rebooted into my FC4 partition with a 2.6.16 kernel, everything much
> >  snappier. So I ran a scp test from my BT server to the laptop, three
> >  times in a row the same file - a 38MB .flac with the laptop in the same
> >  physical position (ie, no signal variation). Results...
> >
> > FC5 - 2.6.16-git3:
> >
> > [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> > Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> > asuardi@192.168.1.8's password:
> > KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB 971.3KB/s   00:40
> > [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> > Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> > asuardi@192.168.1.8's password:
> > KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB   1.3MB/s   00:29
> > [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> > Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> > asuardi@192.168.1.8's password:
> > KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB 626.7KB/s   01:02
> >
> >
> > FC4 - 2.6.16:
> >
> > [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> > Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> > asuardi@192.168.1.8's password:
> > KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB   1.5MB/s   00:25
> > [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> > Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> > asuardi@192.168.1.8's password:
> > KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB   1.7MB/s   00:23
> > [asuardi@donkey melua_2004-09-23_Berlin]$ scp KM_9-23-04_17_The\
> > Closest\ Thing\ to\ Crazy.flac 192.168.1.8:/tmp
> > asuardi@192.168.1.8's password:
> > KM_9-23-04_17_The Closest Thing to Crazy.flac 100%   38MB   1.7MB/s   00:22
> >
> > Bottom line - old driver has better performance than the new one,
> >  but most noticeably delivers consistent performance.
> >
> > I will be available for testing starting Thursday 30th as I'll be on
> >  the road since then. Of course if the problem is identified and
> >  fixed earlier, I won't cry ;)
>
> Well.  It's not a huge regression.  It's a 50%ish regression.  We've done
> worse ;)

That scp test shows 50%ish - but that was a quickie. The VNC
 client even reported a 719Kbps throughput down from the more
 usual 11500Kbps it starts off with. The first scp I tried when the
 sluggishness was intolerable was going at 200KB/s - which
 shows the problem can easily get in the neighborhood of an
 order of magnitude.

Thanks,

--alessandro

 "Dreamer ? Each one of us is a dreamer. We just push it down deep because
   we are repeatedly told that we are not allowed to dream in real life"
     (Reinhold Ziegler)

Re: [2.6.16-gitX] heavy performance regression in ipw2200 wireless driver

[Zhu Yi]

On Thu, 2006-03-23 at 15:02 +0100, Alessandro Suardi wrote:
> That scp test shows 50%ish - but that was a quickie. The VNC
>  client even reported a 719Kbps throughput down from the more
>  usual 11500Kbps it starts off with. The first scp I tried when the
>  sluggishness was intolerable was going at 200KB/s - which
>  shows the problem can easily get in the neighborhood of an
>  order of magnitude.

What kind of wireless encryption do you use? We turned off hardware
encryption by default recently as a workaround for a firmware restart
bug. You might want to load module with "modprobe ipw2200 hwcrypto=1"
and retest.

Thanks,
-yi

Re: [2.6.16-gitX] heavy performance regression in ipw2200 wireless driver

[Alessandro Suardi]

On 3/24/06, Zhu Yi <yi.zhu@intel.com> wrote:
> On Thu, 2006-03-23 at 15:02 +0100, Alessandro Suardi wrote:
> > That scp test shows 50%ish - but that was a quickie. The VNC
> >  client even reported a 719Kbps throughput down from the more
> >  usual 11500Kbps it starts off with. The first scp I tried when the
> >  sluggishness was intolerable was going at 200KB/s - which
> >  shows the problem can easily get in the neighborhood of an
> >  order of magnitude.
>
> What kind of wireless encryption do you use? We turned off hardware
> encryption by default recently as a workaround for a firmware restart
> bug. You might want to load module with "modprobe ipw2200 hwcrypto=1"
> and retest.

I actually use no encryption yet, as I still have to find out time
 to call D-Link about the fact that my router hangs when I try
 to set up a whitelist of MAC addresses for the wireless AP;
 WPA would be up next...

Would loading the module with h/w encryption turned on
 make any difference in my case ?

Thanks,

PS don't tell my neighbors ;)

--alessandro

 "Dreamer ? Each one of us is a dreamer. We just push it down deep because
   we are repeatedly told that we are not allowed to dream in real life"
     (Reinhold Ziegler)

Re: [2.6.16-gitX] heavy performance regression in ipw2200 wireless driver

[Alessandro Suardi]

On 3/24/06, Zhu Yi <yi.zhu@intel.com> wrote:
> On Thu, 2006-03-23 at 15:02 +0100, Alessandro Suardi wrote:
> > That scp test shows 50%ish - but that was a quickie. The VNC
> >  client even reported a 719Kbps throughput down from the more
> >  usual 11500Kbps it starts off with. The first scp I tried when the
> >  sluggishness was intolerable was going at 200KB/s - which
> >  shows the problem can easily get in the neighborhood of an
> >  order of magnitude.
>
> What kind of wireless encryption do you use? We turned off hardware
> encryption by default recently as a workaround for a firmware restart
> bug. You might want to load module with "modprobe ipw2200 hwcrypto=1"
> and retest.

The issue seems to have vanished in more recent kernel snapshots
 (namely, 2.6.16-git3 and -git5 exhibited the problem; -git8 and -git9
 did not).

I will holler if the problem pops up again... thanks,

--alessandro

 "Dreamer ? Each one of us is a dreamer. We just push it down deep because
   we are repeatedly told that we are not allowed to dream in real life"
     (Reinhold Ziegler)